Question about choosing an initial state error covariance for Kalman filter

Hi Everyone,

I have a Kalman filter code from internet, which estimates states of following state space system:

x(k+1)= A x(k) + w(k); y(k)= C x(k);

Before the code starts to perform the Kalman filter, it initializes the initial state error covariance P(0) as

    P{0}= reshape( (eye(n^2)-kron(A,A) ) \ Q(:), n, n );

where n is the size of state, Q is the covariance of white Gaussian noise w(k).

Does anyone have a clue why the code initialize the P(0) like this?

Thanks